Re: [PATCH V2 4/4] ARM64 LPC: support earlycon for UART connected to LPC

[zhichang <zhichang.yuan02@gmail.com>]

On 2016年09月08日 19:04, Arnd Bergmann wrote:
> On Thursday, September 8, 2016 6:04:31 PM CEST zhichang wrote:
>> Hi, Arnd,
>>
>> On 2016年09月07日 22:52, Arnd Bergmann wrote:
>>> On Wednesday, September 7, 2016 9:33:53 PM CEST Zhichang Yuan wrote:
>>>> From: "zhichang.yuan" <yuanzhichang@hisilicon.com>
>>>>
>>>> This patch support the earlycon for UART connected to LPC on Hip06.
>>>> This patch is depended on the LPC driver.
>>>>
>>>> Signed-off-by: zhichang.yuan <yuanzhichang@hisilicon.com>
>>>>
>>>
>>> I'm skeptical about this too. Is this just needed because the 8250
>>> earlycon support comes before the lpc bus initialization?
>> I think you wonder why early_serial8250_setup can not be used direclty for this earlycon of LPC uart.
>>
>> 1. the earlycon kernel parameter format of LPC uart is different from 8250. something like that
>> "earlycon=hisilpcuart,mmio,0xa01b0000,0,0x2f8". You see, there is one more parameter after the baudrate.
> 
> We should never need to specify the addresses manually like this,
> it's actually supposed to work if you just list "earlycon" here.

Do you mean flat-tree earlycon?
Ok, will support this in V3.

> 
> The first membase is apparently only used during setup:
> 
> +       writel(LPC_IRQ_CLEAR, device->port.membase + LPC_REG_IRQ_ST);
> +       /* ensure the LPC is available */
> +       while (!(readl(device->port.membase + LPC_REG_OP_STATUS) &
> +                       LPC_STATUS_IDLE))
> 
> Why doesn't the firmware do this before handing off control of
> the kernel to the console?
This is a checking on the LPC controller status.
I think we can keep this here.

> 
>> Hip06 LPC uart need two base addresses for earlycon.
>> 2. the IO type is mmio to introduce a memory base address to access LPC register file. But the real uart
>> IO type is UPIO_PORT. This is spcial...
> 
> This sounds like a deficiency in the of_setup_earlycon() function,
> which can only handle MMIO addresses, and won't actually
> be able to understand nodes without a "ranges" property like
> you have here.
> 
Yes.
The current of_setup_earlycon only support MMIO and the first reg property must be memory.

We can not support our LPC uart without any new code.
But we can implement a private earlycon setup function and register it to the __earlycon_table, things will be ok.
I will do it in V3.

Best,
Zhichang

> I think we need to add a special case for port ranges here.
> 
> 	Arnd
>